Please increase the number of digits on the leaderboard ! Due to the difficulty to see the improvement of my submissions, I feel like I'm walking through a dark forest. I can't even see the difference to the other competitors….
", "messages": [ { "id": 1324884, "postDate": "2021-05-27T10:09:41.587Z", "content": "Please increase the number of digits on the leaderboard ! Due to the difficulty to see the improvement of my submissions, I feel like I'm walking through a dark forest. I can't even see the difference to the other competitors….
", "rawMarkdown": "Please increase the number of digits on the leaderboard ! Due to the difficulty to see the improvement of my submissions, I feel like I'm walking through a dark forest. I can't even see the difference to the other competitors....\n", "votes": 41 }, { "id": 1324974, "postDate": "2021-05-27T11:54:16.953Z", "content": "We'll be increasing precision a bit in a couple of weeks. But it is kept low currently to avoid probing and overfitting the public leaderboard. The data found in Train is representative enough that local validation improvements will track to improvements in the leaderboard.
", "rawMarkdown": "We'll be increasing precision a bit in a couple of weeks. But it is kept low currently to avoid probing and overfitting the public leaderboard. The data found in Train is representative enough that local validation improvements will track to improvements in the leaderboard.", "votes": 31, "replies": [ { "id": 1325009, "postDate": "2021-05-27T12:33:12.207Z", "content": "That's very good news🙌 Thanks!
", "rawMarkdown": "That's very good news🙌 Thanks!", "votes": 3 } ] }, { "id": 1352963, "postDate": "2021-06-16T18:43:06.673Z", "content": "There we go, competition over?
\nSeveral people found some magic/leak for 0.999?
", "rawMarkdown": "There we go, competition over?\n\nSeveral people found some magic/leak for 0.999?", "votes": 9, "replies": [ { "id": 1352986, "postDate": "2021-06-16T18:59:44.657Z", "content": "It took us several experiments in the last two weeks to reach our current score .986, I do think there is some leak behind achieving .999+
", "rawMarkdown": "It took us several experiments in the last two weeks to reach our current score .986, I do think there is some leak behind achieving .999+ ", "votes": 2 }, { "id": 1352989, "postDate": "2021-06-16T19:01:49.917Z", "content": "It seems to mostly be synthetic data, so I would not be too surprised. 0.999 does not seem doable without finding some leak (I can be wrong, or rather I hope I am), there is also a clear gap between top and rest on lb.
", "rawMarkdown": "It seems to mostly be synthetic data, so I would not be too surprised. 0.999 does not seem doable without finding some leak (I can be wrong, or rather I hope I am), there is also a clear gap between top and rest on lb.", "votes": 2 }, { "id": 1353009, "postDate": "2021-06-16T19:15:41.123Z", "content": "I wonder how many teams would be 0.999 by the competition deadline. Public leaderboard would become useless if many teams has same score.
", "rawMarkdown": "I wonder how many teams would be 0.999 by the competition deadline. Public leaderboard would become useless if many teams has same score. ", "votes": 1 }, { "id": 1353015, "postDate": "2021-06-16T19:17:17.803Z", "content": "The whole competition would be useless, @inversion said 0.995 would already be high but useful. 0.999 seems wrong…
", "rawMarkdown": "The whole competition would be useless, @inversion said 0.995 would already be high but useful. 0.999 seems wrong...", "votes": 1 } ] }, { "id": 1353011, "postDate": "2021-06-16T19:16:13.077Z", "content": "The top 3 , I think the Aliens like you a bit too much . What if everyone finds that leak , the competition will be over soon 😟
", "rawMarkdown": "The top 3 , I think the Aliens like you a bit too much . What if everyone finds that leak , the competition will be over soon 😟", "votes": 5 }, { "id": 1324901, "postDate": "2021-05-27T10:29:33.120Z", "content": "@hirune924 - absolutely agree. Since the LB is about distinguishing between the performance of competing classifiers, it is clearly best practice to quote sufficient precision to demonstrate those differences.
", "rawMarkdown": "@hirune924 - absolutely agree. Since the LB is about distinguishing between the performance of competing classifiers, it is clearly best practice to quote sufficient precision to demonstrate those differences.", "votes": 1 } ], "comments": [ { "id": 1324974, "author_name": "inversion", "author_url": "", "post_date": "2021-05-27T11:54:16.953000", "content": "We'll be increasing precision a bit in a couple of weeks. But it is kept low currently to avoid probing and overfitting the public leaderboard. The data found in Train is representative enough that local validation improvements will track to improvements in the leaderboard.
", "votes": 31, "replies": [ { "id": 1325009, "author_name": "Tawara", "author_url": "", "post_date": "2021-05-27T12:33:12.207000", "content": "That's very good news🙌 Thanks!
", "votes": 3, "replies": [] } ] }, { "id": 1352963, "author_name": "Psi", "author_url": "", "post_date": "2021-06-16T18:43:06.673000", "content": "There we go, competition over?
\nSeveral people found some magic/leak for 0.999?
", "votes": 9, "replies": [ { "id": 1352986, "author_name": "Nischay Dhankhar", "author_url": "", "post_date": "2021-06-16T18:59:44.657000", "content": "It took us several experiments in the last two weeks to reach our current score .986, I do think there is some leak behind achieving .999+
", "votes": 2, "replies": [] }, { "id": 1352989, "author_name": "Psi", "author_url": "", "post_date": "2021-06-16T19:01:49.917000", "content": "It seems to mostly be synthetic data, so I would not be too surprised. 0.999 does not seem doable without finding some leak (I can be wrong, or rather I hope I am), there is also a clear gap between top and rest on lb.
", "votes": 2, "replies": [] }, { "id": 1353009, "author_name": "Nischay Dhankhar", "author_url": "", "post_date": "2021-06-16T19:15:41.123000", "content": "I wonder how many teams would be 0.999 by the competition deadline. Public leaderboard would become useless if many teams has same score.
", "votes": 1, "replies": [] }, { "id": 1353015, "author_name": "Psi", "author_url": "", "post_date": "2021-06-16T19:17:17.803000", "content": "The whole competition would be useless, @inversion said 0.995 would already be high but useful. 0.999 seems wrong…
", "votes": 1, "replies": [] } ] }, { "id": 1353011, "author_name": "Mr_KnowNothing", "author_url": "", "post_date": "2021-06-16T19:16:13.077000", "content": "The top 3 , I think the Aliens like you a bit too much . What if everyone finds that leak , the competition will be over soon 😟
", "votes": 5, "replies": [] }, { "id": 1324901, "author_name": "John Mitchell", "author_url": "", "post_date": "2021-05-27T10:29:33.120000", "content": "@hirune924 - absolutely agree. Since the LB is about distinguishing between the performance of competing classifiers, it is clearly best practice to quote sufficient precision to demonstrate those differences.
", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "1324884": "Please increase the number of digits on the leaderboard ! Due to the difficulty to see the improvement of my submissions, I feel like I'm walking through a dark forest. I can't even see the difference to the other competitors....\n", "1324974": "We'll be increasing precision a bit in a couple of weeks. But it is kept low currently to avoid probing and overfitting the public leaderboard. The data found in Train is representative enough that local validation improvements will track to improvements in the leaderboard.", "1352963": "There we go, competition over?\n\nSeveral people found some magic/leak for 0.999?", "1353011": "The top 3 , I think the Aliens like you a bit too much . What if everyone finds that leak , the competition will be over soon 😟", "1324901": "@hirune924 - absolutely agree. Since the LB is about distinguishing between the performance of competing classifiers, it is clearly best practice to quote sufficient precision to demonstrate those differences." } }